Testing and CI¶

All checks in one command¶

sh tools/check.sh

runs everything a push or a release needs:

1. the tests in tools/test under Node.js 8.12.0, the version webOS 5 and 6 ship (downloaded once from nodejs.org, checksum verified, cached in ~/.cache/cleanyourtv);
2. the same tests under the Node.js of your computer, when it is another version;
3. a check that the launcher and the manager pages are built from their sources (node tools/build-pages.js --check).

With --chromium it also runs the launcher, the manager and the Inputs picker in Chromium 68 and 79, the browser engines of webOS 5 and 6 (tools/oldchrome.js, headless, against a stand-in service): a page that throws a script error there or does not draw its tiles, rows or inputs fails the check. The engines are Chromium snapshots from Google's storage, checked against pinned SHA-256 sums (about 100 MB each, downloaded once). CYTV_SHOTS=<folder> keeps a screenshot of every page per engine; CI keeps them as an artifact of each run.

Nothing leaves your machine apart from the downloads. One area of tests on its own: node tools/test.js live (the names are the files in tools/test without .test.js), or node tools/test/live.test.js.

Types, lint and formatting¶

package.json lists the development tools: TypeScript, ESLint and Prettier. None of them is shipped; the service and the pages have no dependencies. After npm install once, tools/check.sh (or node tools/lint.js) also runs:

• TypeScript on the service (helper/*.js), from the code and its JSDoc comments, without compiling anything (tsconfig.json). It found a missing http import in the voice setup page that no test had reached.
• ESLint on the service, the scripts and the tools, and on the script of every page the TV loads, as ES5 with the TV's globals (eslint.config.js): a name that is used but never defined, a variable declared twice, unreachable code.
• Prettier in check mode. npm run format formats the code and rebuilds the pages.

The tests¶

Each file in tools/test is a small program that prints one line per check (ok or FAIL) and runs in a fresh process with its own temporary folder; helper/paths.js points the service's folders there. Local servers stand in for GitHub, playlists, Home Assistant and the TV's own services.

A fix needs a test that fails without it, whenever the bug can be reproduced without a TV.

Pages¶

The built pages are plain HTML files, so they can be opened in a desktop Chrome with a small mock of the service on 127.0.0.1:8765 and driven with scripted key events (tools/theme-preview.mjs does this for the home screen). That is how the launcher, Live TV and manager screens are checked before they go to a TV. The mock needs to answer /api/apps, /api/live/* and /api/manager/* with the shapes in Service API.

Timings on a TV¶

The launcher, the Inputs picker, the Input key and Live TV write short perf lines to the service log: how long the launcher takes to draw (from the saved layout, and with the service's answer), how long from the Input key to the picker on screen, and how long from choosing a channel to moving picture. After using the TV for a while,

sh tools/perf-tv.sh <tv-ip>

prints them (read-only, over SSH). tools/oldchrome.js prints the launcher's time in Chromium 68 and 79 on your computer, which is only a rough comparison: a TV is slower.

Continuous integration¶

The dev build appears in the manager app's Update channel list; see Dev builds.

Documentation¶

python3 -m venv .venv && . .venv/bin/activate
pip install zensical
zensical serve      # http://localhost:8000, rebuilds on change
zensical build --clean